01 / RESEARCH FRAMEWORK
The proposal
A Promptinal is proposed as a generative artifact whose reconstruction instructions and evidence commitments are associated with a Bitcoin inscription. Instead of assuming the finished asset must be the only durable object, the design asks whether a sufficiently specified recipe can preserve a route to the asset. The recipe may be inscribed directly, or a commitment may identify external material. These choices have different availability, privacy and cost consequences. Neither is a substitute for specifying the complete reconstruction boundary.
02 / RESEARCH FRAMEWORK
Define the artifact before defining the token
The unit of verification must be explicit. For an image, is the target an encoded PNG file, a normalised pixel array or a visually similar scene? Metadata, colour profiles and encoders can change file bytes while leaving an image apparently unchanged. For a model-generated text, exact characters and encoding differ from semantic resemblance. A protocol should name its output format, canonicalisation rule and acceptance test before it promises that two outputs are the same asset.
03 / RESEARCH FRAMEWORK
The generative manifest
A useful manifest commits to a versioned schema, model-weight digests, adapters, tokeniser and other dependencies, input assets, prompt, negative prompt where relevant, seed, sampling parameters, workflow graph and execution profile. It also states the output format and expected cryptographic digest. Human-readable model names are not enough to identify immutable weights. A registry can make a manifest discoverable, but its governance and update policy must not silently change what an existing commitment means.
04 / RESEARCH FRAMEWORK
The Bitcoin record
In the Ordinals model, inscription content is placed in transaction witness data and associated with a satoshi through ordinal indexing. It is not physically stored inside a coin. The research must decide whether the recorded bytes are a complete manifest, a compact reference or another commitment. A network record can establish a particular payload and its transaction history; it does not, by itself, establish who authored an external work or what legal rights were transferred.
05 / RESEARCH FRAMEWORK
Reconstruction as an execution contract
A prompt and seed are insufficient to promise universal byte-identical AI output. The reference environment must specify software versions, numeric precision, hardware assumptions, deterministic operations and serialization. Pinning a WebAssembly runtime may constrain execution, but does not automatically make all model operations deterministic. A serious replay claim needs published inputs and golden outputs, repeated local tests and independent runs across every environment named in the claim. Unsupported environments should produce a clear result rather than a misleading success badge.
06 / RESEARCH FRAMEWORK
Exactness and similarity
A cryptographic digest is useful for testing byte equality with a committed target. A perceptual hash serves a different purpose: it can support a similarity comparison, but a similar image is not necessarily the exact same file. The recovered drafts discuss perceptual hashing alongside deterministic generation. This edition separates those ideas. A specification can define both a strict byte check and a separate similarity measure, provided the interface never labels one as the other.
07 / RESEARCH FRAMEWORK
Availability is a separate promise
External model references reduce the payload placed in an inscription only by leaving some information elsewhere. A content address identifies expected content; it does not guarantee that a provider will continue serving it. A reconstruction plan therefore needs retention, replication, retrieval and integrity checks for model weights and inputs. Redistribution rights matter too: a reproducibility design cannot assume every dependency may be mirrored publicly. Storage, computation and verification costs should be measured together.
08 / RESEARCH FRAMEWORK
Access, signatures and ownership
The early access-gated route proposes checking control before returning an off-chain prompt. That can describe a service policy, but a public satoshi identifier is not a secret encryption key. Once a recipient can copy a public recipe or decrypted content, transfer of a token cannot make the copy disappear. Signed manifests can attribute statements to keys, and inscription history can support provenance. Authorship, copyright, licensing and contractual rights require their own evidence and terms.
09 / RESEARCH FRAMEWORK
Distributed compute and the proposed overlay
The wider research connects resource sharing with generative workloads. A distributed worker network would need job specification, input availability, output verification, failure handling, resource limits and an incentive design that remains credible under adversarial behaviour. PLEB—Promptinals Ledger of Extensibility for Bitcoin—is described in drafts as a proposed overlay for asset state and extensibility. Its state-transition rules, proofs and trust assumptions require specification. Anchoring a claim to Bitcoin does not automatically validate the computation or overlay that produced it.
10 / RESEARCH FRAMEWORK
A programme of validation
Begin with a minimal manifest and one narrowly supported execution profile. Publish test vectors and a reference implementation, then ask independent implementers to reproduce them. Test missing models, altered dependencies, corrupted outputs, replay across versions and unavailable storage. Define reorganisation handling and inscription parsing separately from execution verification. Add security review before operating a registry, access service or distributed-worker system. Publish measured outcomes and failed cases alongside successes. The public SVG lab demonstrates only a small deterministic verification pattern.
11 / RESEARCH FRAMEWORK
Current status and scope
This is a newly authored research synthesis, not a verbatim copy or final release of a recovered white paper. The sources establish documented concepts, draft architectures and contemporary gap reviews. They do not establish an infallible system, production PLEB network, independently audited protocol or cross-platform AI replay result. Historical roadmaps are proposals from their time, not proof of completion. The useful next step is an inspectable specification and reproducible evidence.